Professional Overview
PredictAI, Independent Project (Python, FastAPI, React, time-series ML, Ollama) 2026 • Built an offline desktop app that turns a shop's sales file into "order this many" for every product, with stockout and dead stock alerts, for non-technical pharmacy and retail owners in the Gulf (Arabic/English); 650+ automated tests. • Cut forecast time 80% (203 s to 39 s) by benchmarking models on 9 real datasets and running the slow models only where they lowered error. • Grew coverage from 20 products to all products (up to 1,000); before, a 60-product pharmacy could act on only 77% of its stock. • Per-product model selection by backtesting across statistical models (AutoARIMA, ETS, Theta, Croston/TSB) and AI foundation models (Chronos, TimesFM), with confidence ranges, plus a private on-device AI analyst (Qwen 2.5). LLM Fine-Tuning Research (PyTorch, Unsloth, vLLM, Hugging Face, JAX/Flax) • Implemented a GPT-style language model from scratch in JAX/Flax to learn transformer training end to end. • Fine-tuned Qwen2.5-7B-Instruct with LoRA/QLoRA adapters (SFT) and GRPO reinforcement learning using PyTorch, Unsloth and Hugging Face, with vLLM for fast inference during evaluation.
Areas of Expertise
AI & Machine Learning Engineers
Design, train, and ship production ML models — from data pipelines to model inference at scale. Covers supervised learning, neural networks, LLM integration, and model optimisation.
Data Science & Analytics
Extract business insights from raw data using statistical modelling, predictive analytics, A/B testing, and interactive dashboards.
NLP & Prompt Engineering
Build natural language processing systems, fine-tune large language models, and craft reliable production prompts for AI-powered products.
